NEWS

Anthropic Rents All of SpaceX’s Colossus 1 Data Center

Anthropic SpaceX Colossus 1 data center compute deal in Memphis
0:00
0:00🎧 6 min

MEMPHIS, Tenn.: Anthropic has signed a compute partnership with SpaceX that gives it exclusive access to all of the Colossus 1 data center in Memphis, the companies confirmed in a joint announcement, putting more than 220,000 Nvidia GPUs and over 300 megawatts of AI capacity under Anthropic’s control. The deal, disclosed by Anthropic in a company post on higher Claude limits and confirmed by xAI in a compute partnership announcement, hands the Claude maker a chunk of infrastructure originally associated with Elon Musk’s xAI stack.

The companies said the GPU mix at Colossus 1 spans H100, H200, and next-generation GB200 systems. Anthropic plans to route that capacity to paying Claude customers immediately, with the first concrete effect being higher rate limits on Claude Code and the Opus API.

How the Anthropic SpaceX deal reshapes AI infrastructure supply

Colossus 1 was originally built and operated by xAI as one of the largest single-site AI training clusters in the United States. By renting the full data center to Anthropic, SpaceX effectively converts an in-house training asset into a revenue-generating capacity contract with a direct competitor. Musk addressed the optics himself on X, saying “No one set off my evil detector” when asked why SpaceX would supply a top rival.

The size of the lease matters. Three hundred megawatts is roughly the power draw of a mid-size city and sits at the top end of contiguous AI compute available on the open market. Few other operators can offer a single tenant that much capacity in one location, which is part of why Anthropic was willing to source from a competitor’s stack rather than wait for new builds from its existing cloud partners. The squeeze is industry-wide: Google’s cloud backlog ballooned past $462 billion in Q1 2026, a sign of how thin available capacity has become for the labs racing to ship next-generation models, as detailed in our coverage of Google Cloud’s $462B backlog.

The financial backdrop sharpens the picture. Anthropic is weighing a $50 billion funding round at a valuation in the $850 billion to $900 billion range, according to TechCrunch reporting. That would more than double the $380 billion valuation set in February, when Anthropic last raised alongside its expanded AWS partnership with Amazon. The company also disclosed an annual revenue run rate north of $30 billion, up from roughly $9 billion at the end of 2025. Capacity, not capital, is the binding constraint.

What does the Anthropic Colossus data center deal mean for founders building on Claude?

For founders shipping products on Claude, the deal converts directly into higher throughput. Anthropic doubled Claude Code’s five-hour rate limits for Pro and Max tiers, removed the peak-hour limit reduction on those tiers, and “considerably” raised API rate limits for Claude Opus, the company said in its announcement. Agentic workflows that previously hit ceilings during business hours should now run closer to peak.

The broader signal for builders is that frontier labs are competing on raw infrastructure access, not just model architecture. A year ago, model choice was the main lever. Today, a lab’s ability to secure 300-megawatt blocks of GPUs determines what kind of product behavior it can offer at scale. Founders should expect rate limit changes, pricing changes, and feature gating to track compute deals as closely as model releases.

It also reframes vendor risk. The line between supplier and competitor is now porous: SpaceX is leasing capacity to Anthropic while xAI ships Grok, and Amazon is both a major Anthropic investor and an operator of competing Bedrock models. Geographic risk is part of the equation too. Maine’s new data center moratorium showed that state-level decisions can close off compute footprints overnight. Treat any single-vendor or single-site dependency as a relationship that can shift quarter to quarter.

What’s next for Anthropic, SpaceX, and orbital compute

Anthropic and SpaceX said in their announcements that they are exploring a partnership to develop multiple gigawatts of orbital AI compute. The companies have not committed to a timeline or capital figure, and the technical, regulatory, and thermal questions around space-based data centers remain open. Power delivery, radiation hardening of accelerators, and downlink bandwidth all need credible answers before orbital training becomes more than a press-release ambition. Even framed as exploratory, the statement puts orbital compute on the near-term roadmap of a frontier lab rather than in research papers.

The closer milestone to watch is Anthropic’s funding decision. A board meeting in May is expected to finalize the $50 billion round, with a public listing possible as early as October 2026, per TechCrunch sourcing. Founders relying on Claude should also watch for Anthropic’s next rate-limit and pricing updates, which will be the clearest read on how the new Colossus capacity is being allocated across paid tiers, enterprise contracts, and internal model training.

Read More From the NEWS desk